Blog

Writing & research.

Benchmarks, research and the engineering behind African-language and code-switch dictation.

Benchmark·

Sema Ink v1 vs the GhanaNLP Akan benchmark

We ran our code-switch model on a benchmark it was never trained on. At 49.5% WER it beats every other out-of-domain model, and most of the error turns out to be dialect, not mistakes.

Read →

Coming soon

The first Twi / Pidgin / English code-switch benchmarkSoon
One CPU box to one GPUSoon
Does code-switching transfer?Soon
float16 beats int8Soon
What fine-tuning Whisper on Twi does to its other 98 languagesSoon
The correction flywheelSoon
← Back home